Micro-motion Classification of Rotor UAV and Flying Bird via CNN and FMCW Radar

Abstract

Aiming at the problem that it is difficult to recognize flying birds and rotary-wing UAVs by radar, a micro-motion feature classification method based on multi-scale convolutional neural network (CNN) is proposed in this paper. Using the K-band frequency modulated continuous wave (FMCW) radar, data acquisition is performed on the rotor UAV and flying bird targets in indoor and outdoor scenes, and then the feature extraction and parameterization of the micro-Doppler signal are performed using time-frequency analysis technology to construct the radar feature dataset. A novel type of multi-scale CNN is designed, which can extract the global and local information of the target’s micro-Doppler features and improve the classification accuracy. Validation of measured data shows that the classification probability of rotary-wing drones and flying bird targets can reach higher than 98% by using the proposed algorithm, which provides a new technical and practical approach for the identification of low and slow small targets.